Quick-Start

Papers:

• RFC 4782: Quick-Start for TCP and IP.
  Sally Floyd, Mark Allman, Amit Jain, and Pasi Sarolahti.
  RFC 4782, Experimental, January 2007. Errata.
  Local copy: (postscript, text). Quick-Start web page.

• RFC 5634: Quick-Start for the Datagram Congestion Control Protocol (DCCP).
  G. Fairhurst and A. Sathiaseelan.
  RFC 5634, Experimental, August 2009. DCCP web page.

• Determining an Appropriate Sending Rate Over an Underutilized Network Path.
  Pasi Sarolahti, Mark Allman, and Sally Floyd.
  Computer Networks, V.51 N.7, May 2007. Simulation scripts.

• Using Quickstart to Improve the Performance of TFRC-SP over Satellite Networks.
  Arjuna Sathiaseelan and Gorry Fairhurst,
  IEEE/IWSSC 2006, Madrid.

Internet drafts:

• TCP Flow Control for Fast Startup Schemes,
  M. Scharf, S. Floyd, and P. Sarolahti,
  Internet-draft draft-scharf-tcpm-flow-control-quick-start-00.txt, work-in-progress, July 2008.
  
  Was: Avoiding Interactions of Quick-Start TCP and Flow Control,
  Internet-draft draft-scharf-tsvwg-quick-start-flow-control-01.txt.

Implementations:

• Experiences with Implementing Quick-Start in the Linux Kernel,
  M. Scharf and H.. Strotbek, July 2007, IETF.

Talks:

• Quick-Start for TCP and IP: talks by Sally at TSWWG at IETF.
  TSVWG, IETF, March 2006 (PDF, ppt).
  TSVWG, IETF, November 2005 (PDF, ppt).
  TSVWG, IETF, August 2005 (PDF, ppt).
  TSVWG, IETF, March 2005 (PDF, ppt).
  TSVWG, IETF, November 2004 (PDF, ppt).
  TSVWG, IETF, March 2003 (postscript, PDF).
  TSVWG, IETF, July 2002 (postscript, PDF).

• Sally Floyd, Congestion Control for High-Bandwidth-Delay-Product Networks: XCP vs. HighSpeed TCP and QuickStart,
  ICIR Wednesday Lunch, September 11, 2002 (postscript, PDF).

Simulations:

Srikanth Sundarrajan added QuickStart to the NS simulator in 2002. The initial validation test is "./test-all-quickstart" in tcl/test.
Pasi Sarolahti updated the Quick-Start implementation in September 2004, to conform to draft-amit-quick-start-03.txt.

The default values for the Quick-Start parameters in ns-default.tcl were updated on 2/25/2005. With the NS defaults from 2/25/2005, the router approves Quick-Start requests when the link utilization will remain below 85%. The router measures the current link utilization by measuring the utilization over five 200-ms time slots, and taking the peak measurement.

Open Questions:

• What do measurement studies show about underutilized end-to-end paths, where QuickStart would be of the most benefit? Are there a lot of underutilized paths? Only in Intranets? Or only for not-at-home users with very good Internet connectivity? Are there likely to be more underutilized paths in the future?

• What are the range of options for more precise algorithms in routers, for those routers that are willing to do a little more work, or to keep a little more state?

Other Proposals for Positive Feedback from Routers:

• S. Kunniyur, AntiECN Marking: A Marking Scheme for High Bandwidth Delay Connections, to appear in Proceedings, ICC, May 2003.
  "We describe a simple scheme that uses feedback from underutilized high capacity links to allow a TCP connection to aggressively increase its sending rate. The feedback is in the form of a single bit in the packet header and is given per packet."

• XCP (eXplicit Control Protocol). (In particular, Congestion Control for High Bandwidth-Delay Product Networks,
  Dina Katabi, Mark Handley, and Charlie Rohrs, Sigcomm 2002.)

(Some of these proposals are rather different from QuickStart, in that they allow more precise control and faster ramping-up than TCP, but they don't enable a large initial window.)